FIG. 1 illustrates a cryptologically weak random number generator (CWRNG) 100. Examples of CWRNG's are a linear congruential random number generator (LRNG) and a quadratic congruential random number generator (QRNG). Random number generators are widely used in cryptographic applications, such as for encrypting digital computer messages transmitted over a network.
LRNG's are recursions in x of a form xi+1=(a*xi+b)%c, where * denotes multiplication and % denotes a modulo function. For example, (a*xi+b)%c would produce an integer in a range of 0 to (c-1). a, b and c, the LRNG parameters, along with x0, the LRNG starting value determine succeeding values of x. The LRNG will cycle through all numbers in the range from 0 to c-1 in pseudo-random order before repeating (i.e. it has a period of c, the maximum period), if and only if:    1. b is relatively prime to c (that is , no prime factors of c are prime factors of b);    2. (a-1) is a multiple of p for every prime number p dividing c; and    3. (a-1) is a multiple of 4 if c is a multiple of 4.
Further, in order to prevent (a-1) from being a predictable value, c should not divide evenly into (a-1). This can be accomplished by imposing a fourth rule as follows:    4. at least one prime factor in c is raised to a power greater than 1 and the same prime factor used in (a-1) must be raised to a power less than the power used to raise that prime factor in c.
QRNG's are recursions in x of a form x=(d*x*x+a*x+b)%c, which generates a series of integers in a range from 0 to (c-1). a, b, c and d, the QRNG parameters, along with x0, the QRNG starting value determine the succeeding values of x. A QRNG will cycle through all numbers in a range of 0 to (c-1) in psuedo-random order before repeating (i.e. it has a period of c, the maximum period), if and only if:    1. b is relatively prime to c;    2. d and (a-1) are both multiples of p for every odd prime number p dividing c;    3. d is even and d is congruent to (a-1)%4 if c is a multiple of 4;    4. d is congruent to (a-1)%2 if c is a multiple of 2; and    5. either d%9=0 or a%9=1 and (b*d)%9=6 if c%9=0.
Further, in order to prevent (a-1) and d from having predictable values, c should not divide evenly into (a-1) and d.
While LRNG's and QRNG's provide a fast and efficient method for generating quasi-random numbers, they have certain disadvantages. An outside observer can determine succeeding outputs from either a LRNG or a QNRG after observing a short sequence of outputs.
FIG. 2 shows a two-stage random number generator, which includes a cryptologically weak random number generator 100 with an output stream 110. The output stream may be encrypted by a block encryptor 210 making it difficult for an outside observer to successfully attack this system by observing the output 212 of the block encryptor 210.
Random number generators may be used in a communication system (e.g., Internet) where large numbers of random number generators may be used and frequently re-keyed to provide security to communications occurring in the system by, for example, providing security keys for encrypting and decrypting such communications.
Data in a communication system is placed in packets. Each packet is sent from a source to a destination and may pass through one or more intermediate locations before reaching its destination. The security keys allow for the data to be encrypted at the source and decrypted at the destination, such that an unintended recipient, for example, a hacker monitoring one of the intermediate locations, is unable to decrypt the data contained in the packets because he does not possess the security keys.
Random numbers generated by cryptologically weak random number generators will eventually repeat. The repeating numbers are in an easily guessable sequence of random numbers. The cryptologically weak random number generator, the encryptor or both, should be re-keyed well before the sequence repeats.
The time for re-keying the cryptologically weak random number generator should be short for two reasons: (1) a system crash will make the re-keying of many cryptologically weak random number generators necessary and slow re-keying would effectively extend an outage; and (2) re-keying is non-productive overhead for a system during normal “sunny day” operation of the system. As such, it is appropriate to limit this overhead to be at most x percent of the system's resources. The overhead is limited to x percent of the system's resources when:T<=0.01x c/p,  (1)where T is the time to re-key in seconds, c is the number of packets transmitted between re-keys (directly proportional to cycle length) and p is the maximum speed of the data in packets per second. For instance, if x=0.1, c=1,000,000 packets, and p=100,000 packets per second, then T can be at most 10 milliseconds, with re-keying occurring every 10 milliseconds. Typically, cryptologically weak random number generators, such as linear congruential random number generators (LRNG's) and quadratic congruential random number generators (QRNG's) can be combined to achieve a long cycle. Therefore, re-keying after a crash is a limiting constraint.
The re-keying should yield practically unguessable random number generators. The parameters for the random number generator should not exhibit a strong bias that would make them easily determined by an observer. For instance, if the parameters for a ten parameter cryptologically weak random number generator were restricted to being one of the 6,542 primes between 2 and 65,535, over ten orders of magnitude fewer unique cryptologically weak random number generators would be possible than would be if the parameters were selected as being any integer in the range 2-65,535. However, generating cryptologically weak random number generators with known long non-repeating cycles restricts the choice of parameters and increases the predictability of parameters. This increase in predictability should be minimized.